RFC: fp printing speedup patch

[Martin Sebor]

Paolo Carlini wrote:

...
> My point was that a few virtul function calls doing nop should (and in 
> fact if
> you benchmark _M_insert_float this is the case) not cost more that one
> function call doing memcopy inside, however.

FYI, the cost of a virtual function is architecture dependent. On
HP PA, for instance, virtual function calls are extremely expensive
(about 5 times more than ordinary function calls), while on HP IPF
and SPARC they are only about twice as expensive.

Martin